LLM Server with Ollama and MCP Support

A comprehensive LLM server that integrates Ollama for local language model inference with Model Context Protocol (MCP) support for tool usage, including advanced geospatial analysis.

Features

• Ollama Integration: Connect to local Ollama models
• MCP Support: Built-in Model Context Protocol for tool integration
• OpenAI-Compatible API: Standard chat completions endpoint
• Streaming Support: Real-time response streaming
• Geospatial Analysis Tools: List, analyze, and visualize geospatial layers
• Built-in Tools: File operations, web search, calculator, and more
• FastAPI Backend: High-performance async API server

Prerequisites

1. Install Ollama: Download and install from ollama.ai
2. Pull Models: Download models you want to use

   ollama pull llama2
   ollama pull mistral
   ollama pull codellama

Installation

1. Install dependencies:

   pip install -r requirements.txt

2. Start the server:

   python llm_server.py

The server will start on http://localhost:8000

API Endpoints

Chat Completions

POST /chat/completions

OpenAI-compatible endpoint for chat completions, with tool support.

Example Request:

{
  "model": "llama2",
  "messages": [
    {"role": "user", "content": "What layers are available?"}
  ],
  "tools": true,
  "stream": false
}

List Models

GET /models

Returns available Ollama models.

List Tools

GET /tools

Returns available MCP tools (including geospatial tools).

Execute Tool

POST /mcp/execute_tool?tool_name=<tool_name>

Execute an MCP tool directly.

Health Check

GET /health

Server health status.

Built-in MCP Tools

General Tools

• read_file: Read file contents
• web_search: Search the web using DuckDuckGo
• get_page_content: Fetch and extract text content from any web page
• validate_url: Check if a URL is accessible and get basic information
• calculate: Perform mathematical calculations

Geospatial Tools

• list_layers: List all available geospatial layers
• find_layer_by_name: Find a layer by its name
• get_layer_info: Get detailed information about a specific layer
• analyze_population: Analyze population data in a layer
• filter_features: Filter features in a layer by attribute conditions
• get_attribute_stats: Get statistics for a specific attribute in a layer
• spatial_analysis: Perform spatial operations (buffer, intersection, etc.)
• create_map_visualization: Create a map visualization with styled features
• analyze_layer_attributes: Comprehensive attribute analysis for a layer

Enhanced Web Search Features

The web search tool now provides real browser-based search capabilities:

• DuckDuckGo Integration: Uses both API and HTML scraping for comprehensive results
• Instant Answers: Gets direct answers for factual queries
• Definitions: Provides definitions for terms and concepts
• Related Topics: Suggests related information and links
• Fallback Search: HTML scraping when API doesn't return results
• Clean Results: Properly formatted with titles, snippets, and URLs

Web Search Examples

# Basic web search
{
  "model": "llama2",
  "messages": [
    {"role": "user", "content": "Search for information about quantum computing"}
  ],
  "tools": ["web_search"]
}

# Get page content from a specific URL
{
  "model": "llama2", 
  "messages": [
    {"role": "user", "content": "Get the content from https://example.com/article"}
  ],
  "tools": ["get_page_content"]
}

# Validate a URL before accessing
{
  "model": "llama2",
  "messages": [
    {"role": "user", "content": "Check if https://example.com is accessible"}
  ],
  "tools": ["validate_url"]
}

Usage Examples

Basic Chat

import requests

response = requests.post("http://localhost:8000/chat/completions", json={
    "model": "llama2",
    "messages": [
        {"role": "user", "content": "What is the capital of France?"}
    ]
})

print(response.json())

Geospatial Analysis Example

response = requests.post("http://localhost:8000/chat/completions", json={
    "model": "llama2",
    "messages": [
        {"role": "user", "content": "Analyze the attributes of the tl 2024 36 place layer"}
    ],
    "tools": true
})
print(response.json())

Direct Tool Execution

response = requests.post(
    "http://localhost:8000/mcp/execute_tool?tool_name=list_layers",
    json={}
)
print(response.json())

Extending with Custom Tools

You can add custom MCP tools by extending the MCPServer class in llm_server.py. See the code for examples.

Configuration

The server runs on:

• Host: 0.0.0.0 (all interfaces)
• Port: 8000
• Reload: Enabled in development mode

Modify the uvicorn.run() call in `